Gather more data with automatic pickingΒΆ

The ab initio protocol has successfully generated a reference, which we will use for the refinement process. Since the refinement method is significantly less computationally intensive with respect to particle count, we can work with a much larger number of particles compared to the ab initio approach. To achieve this, in this section, we will use an automatic picking pipeline to provide the refinement algorithm with as many particles as possible.

The automatic picking pipeline:

  • Train the neural network

    Use the picking train protocol to train a neural network. A key parameter here is the number of particles per image, which we can set to 50 in our case.

  • Predict coordinates

    After training, use the picking predict protocol to apply the trained neural network to a set of images. This step generates several sets of coordinates. Only the output named output3DCoordinates_last_step is relevant for our needs.

  • Extract particles

    Finally, use the extract particles protocol to extract a set of 3D particle images based on the predicted coordinates.

Important

Automatic picking can result in false positives. Make sure to filter these out using the select subset protocol.

Note

If you were unable to run the picking pipeline, pre-picked particles are available in the data/real/ISIM/cropped directory. You can import these using the import particles protocol.